Three reversible controls for Spotify’s Recently Played shelf. Each one completes in one or two taps.
Self-directed concept. Recently Played is one of Spotify’s highest-traffic surfaces and one of its least controllable. This case study designs three additive, reversible controls without rebuilding the shelf or breaking the recommendation engine.
For power users · additive · reversible
Spotify’s Recently Played is one of the best surfaces in consumer music. For power users on shared screens, family devices, and one-off plays, three lightweight controls add the lever they keep reaching for. The shelf broadcasts your recent plays to anyone sharing the screen, and power users are the ones who notice and care. All reversible, all native to the shelf, none touch the recommendation engine.
“Workarounds documented in community forums are signal. People are reaching for affordances that don’t exist yet.”
Contextual friction
Four predictable pain points that show up when Recently Played doesn’t behave the way the user’s context expects.
Recently Played sits at the top of every Home feed and updates automatically from listening history. Always visible, on your screen and on anyone else’s. The shelf records every play and orders by recency. That’s the entire interaction model. Workarounds exist: community forums document users playing a track for one second and force-quitting to push it to the top. That behavior is signal. People are reaching for affordances that don’t exist yet.
No internal data, no brief. The research layer was built from public sources: Spotify Community threads, App Store reviews, Reddit, UX forum posts. AI-assisted clustering on 200+ posts surfaced six quotes that made the design unavoidable.
“Recently Played is literally the ONE AND ONLY feature of the app I use.”
Reddit, Oct 2024
“I'd play 1 second of a track so it would show in Recently Played.”
Spotify Community, 2020
“That annoyed me so much I left Spotify.”
Reddit, Jul 2023
“Listening history is like a diary of my music journey.”
Spotify Community, 2021
“Apple Music lets you clear listening history. Spotify should too.”
App Store, 2023
“Why hasn't Spotify added a clear history option yet?”
Spotify Community, Jun 2020
Primary persona
Melodic Melanie
The social-pressure case. Pin to surface favorites she'd share, Remove to clear plays she wouldn't.
Secondary persona
Ranger Dave
The time-pressure case. 45 minutes on BART. If a control takes more than two steps, he won't use it.
Negative persona
Stephen Tan
The boundary case. He doesn't engage with the shelf. If the design started serving Stephen, the feature would be too broad.
Heuristic audit across seven listening platforms. Apple Music, Amazon Music, and YouTube Music all ship at least one shelf-level control. TikTok and YouTube treat the feed as inline-editable by default. Spotify ships none. The pattern is in the open.
The five categories
Clear all · Pause history · Remove single · Inline controls · Reversible.
| Platform | Clear all | Pause history | Remove single | Inline controls | Reversible |
|---|---|---|---|---|---|
| YouTube Music | ✓ | — | — | Limited | ✓ |
| Apple Music | — | ✓ | ✓ | — | ✓ |
| Amazon Music | — | ✓ | ✓ | ✓ | ✓ |
| TikTok / YouTube | n/a | n/a | n/a | ✓ | ✓ |
| Spotify (today) | — | — | — | — | — |
| Spotify (proposed) | ✓ | ✓ | ✓ | ✓ | ✓ |
Swipe to see more →
Spotify today: no per-item control of any kind. The proposed row shows the three new controls (plus a settings-level Clear All, scoped out of this case study).
I started with three controls for the Recently Played shelf: Pin, Remove, and Pause. I couldn’t ship all three as equals, so I ranked them against one question. Which one fixes a problem the shelf actually leaves you stuck with?
Remove won. The shelf shows what you played to anyone looking at your screen, and there’s no clean way to take something off it. Pin is additive, a preference, not a fix. Remove solves the actual pain.
Pause stayed, but time-boxed. A permanent pause changes what the shelf is. A temporary one hands you control for the moment you need it, then the shelf returns to normal on its own.
Remove has to be reversible. If it deletes instead of hides, it repeats the exact destructive pattern I’m critiquing, so undo is part of the control, not an add-on.
Pause needs an expiry you can see. When the time-box ends, the shelf comes back on its own, and no one should have to remember they paused it.
One-frame journey across the Recently Played shelf. Peaks and valleys mapped against the three opportunity points where Pin, Remove, and Pause each earn their slot.
Swipe to explore · full view on desktop →
Three interaction directions sketched on paper before any Figma frame. The winning direction survived three tests: does it preserve discovery, is it reversible, does it sit native to the shelf. Mid-fi mocked the answer on both surfaces (mobile + desktop) before any hi-fi work.
Sketches
Each direction tests a different affordance for the same shelf. Lightweight, reversible, native to the surface the control lives on.
pin
Right-click a tile. Pin on top sits at the top of the menu, with the four-pin cap named inline as a header.
pause
A shelf-scoped pill above Recently Played. One click freezes the whole shelf, and the state stays visible.
remove
Long-press opens the action sheet with Remove on the surface. Don't suggest similar sits below for the softer signal.
Mid-fi · desktop + mobile
The wireframes the annotated walkthrough below refers to. Worth seeing the layouts raw before reading the rationale behind each interaction beat.
Each control gets the same editorial treatment: title, the annotated walkthrough, then three paragraphs naming the problem, the trade-off, and who wins. The looping prototypes live in their own tabbed section below.
PIN · Mobile
Max four pins. Long-press, choose, swap. The cap teaches itself.
Problem and state coverage. Recently Played reorders by recency. A song the user wants to keep within reach disappears as new plays push it down. The shelf needs a way to lock a favorite in place. The new Pin state machine covers seven moments: default, hover, press, pinned, disabled (cap reached), loading, error. Each has two independent signals so the affordance never reads as ambiguous.
Feedback and the cap. Move first, confirm second. The tile animates to position one in 250 ms; the Pinned. Undo snackbar follows for five seconds; a persistent Pin chip stays on the tile after the snackbar dismisses. The cap is four pins, with a disabled fifth row rather than a soft limit and a warning toast. Engineering surfaced shelf-performance and ranking-signal degradation past four; four is also the cognitive ceiling for what a user can re-find by sight. Drag-to-reorder among pins, deliberately cut.
Persona impact. Ranger Dave on his commute is the primary beneficiary. Pin keeps his focus playlist within one tap regardless of what he plays after. Parent is neutral. Pin on a shared device is rarely worth doing. Explorer is neutral to negative; heavy browsing means the shelf turns over quickly and four pins barely scratch the volume.
REMOVE · Mobile
Per-device. Five-second undo. Library untouched.
Problem and state coverage. An accidental tap, a short preview, a one-off experiment lingers on the shelf and misrepresents what the user cares about. They need a way to clean it up before someone sees. The Remove state machine handles default, hover, press, the successful remove, the active toast, sync conflicts, and the restored state after Undo. Single-item rows and multi-item rows handled distinctly.
Feedback and scope. Slide first, undo second. The tile slides out in 220 ms. The Removed. Undo snackbar runs for five seconds. On undo, the item returns to its last position rather than the front. Scope is per-device, not per-account. Privacy lives at the device boundary. The cost is friction for multi-device users. The benefit is that the shelf can’t betray a user’s intent on Device A by changing on Device B.
Persona impact. Melodic Melanie, the primary beneficiary, can clear contextual plays before she shares her screen. Ranger Dave is positive on the shared-device case. Explorer is neutral. The shelf already turns over fast for them.
PAUSE · Mobile
Time-boxed. 30 minutes, 2 hours, until tomorrow. Resume is automatic.
Problem and state coverage. A borrowed phone, a long drive with a guest, an overnight kid takeover. The user has no way to stop the shelf from logging without leaving the app or signing out. The Pause state machine surfaces the timer in every state: default, hover, press, paused, expiring, resumed, clock-skew error. The timer is the design’s promise.
Feedback and the time-box. Three presets, no slider. The duration picker offers 30 minutes, 2 hours, or until tomorrow. The shelf shows the paused state with a countdown; a snackbar confirms when listening history resumes. The control is time-boxed only, not a permanent toggle. Permanent pause was ruled out because the ML signal would degrade. The cost: a user who really wants to opt out has to re-pause periodically, or use Settings · Privacy.
Persona impact. Parent is the primary beneficiary: Pause protects recommendations during somebody else’s listening session. Melodic Melanie is positive on the social case. Ranger Dave is positive on shared devices when his son’s music threatens to take over the shelf.
Interactive prototype · try the three controls
Pick a row in the shelf below, then use one of the three controls under it. The demo runs against the constants above.
PIN_CAP = 4 · UNDO_WINDOW = 3s · PAUSE_PRESETS = [30m, 2h, tomorrow]
Three additive controls, all native to the shelf, each completing in one or two taps. Pin to keep a favorite within reach. Remove to hide an accidental or contextual listen without deleting it from the library. Pause to stop the shelf from logging for a time-boxed session. Reversible by default. Nothing touches the recommendation engine. Permanent pause was ruled out because the ML signal would degrade. Time-boxing is what survived the stakeholder map.
Three controls, two affordances (mobile long-press, desktop right-click), one state machine. Below: the Pause state diagram that locks the time-boxed behavior, then the honest scope of what wasn’t tested.
State diagram · Pause
Swipe to see more →
Out of scope
Knowing what not to design is the same skill as knowing what to design. Each item below was raised, considered, and deliberately left for a different surface or a different team.
Skipped
Editing global listening history
This touches a different surface (Settings → Privacy) and a different stakeholder group. Scoping it here would expand the project beyond one sprint.
Skipped
Changing ranking algorithms
Recommendation logic is owned by a separate team. Any ranking changes would require cross-team alignment that's out of scope for a shelf-level UX feature.
Skipped
Profile privacy redesign
Friends-feed and listening visibility live in a different settings surface. They deserve their own design pass, not a piggyback on this one.
Skipped
Permanent pause toggle
Permanent pause was ruled out at the constraint stage. The ML training signal degrades fast with persistent pause. Time-boxing is what survived.
Mobile uses long-press to open the action sheet. Desktop uses a right-click context menu. The three controls and their state machines are identical. Only the affordance changes.
Swipe to see more →
This is a concept, so there are no live numbers and I won’t invent any. Here’s the test I’d run instead. Put the prototype in front of five power users who share a screen at home or at work. Success is narrow: they can take an item off the shelf in one action, undo it if they change their mind, and say what Pause does without me explaining it. If they can’t undo, Remove is just the destructive pattern again, and the concept fails its own point.
Design log · the receipts
First stakeholder read of the brief. Framed the problem as agency without signal loss, not history editing.
Ruled out permanent pause. Time-boxing survived the ML-signal-integrity review.
Cap set at four pins after engineering flagged shelf-perf degradation and ranking-signal noise past four.
Undo window locked at five seconds. Matches the existing snackbar pattern; nothing new for the user to learn.
Mobile long-press vs desktop right-click parity confirmed. Affordance changes, state machine does not.
Reduced-motion path documented. Every animated state has a static fallback that carries the same information.
Signed · Alfonso Barreiro
